Negative Control Outcomes to Assess the Comparability of Biologics versus Small-molecule Therapeutics for Osteoporosis: Considerations for Differential Non-adherence

Background: In a comparative new user study, censoring for non-adherence attempts to estimate the effect of initiating and remaining on treatment. A growing focus on biologics across many therapeutic areas raises questions about unmeasured adherence mechanisms among users of these injectable therapies versus oral small-molecule drugs, which are typically less costly, dosed at more frequent intervals, and unlikely to require administration by a medical professional. Negative control outcomes (NCOs) can be used to detect residual bias when differential adherence is suspected.

Objectives: We used NCOs to assess the comparability of women receiving biannual denosumab injections to those receiving once-weekly oral alendronate for postmenopausal osteoporosis (OP). To gauge the impact of differential adherence, we compared results from an intention-to-treat (ITT) analysis to an analysis that censored for treatment discontinuation.

Methods: We identified women aged ≥55 years in the 2010-2019 de-identified Optum© Clinformatics® Data Mart Database who switched to denosumab (N=11,832) or alendronate (N= 8,658) after receiving another antiresportive in the prior 455 days. We assessed the potential for residual bias when comparing denosumab and alendronate by estimating 12-month cumulative risk differences (RDs) for 12 NCOs. RDs were estimated using a doubly robust augmented inverse-probability of treatment and censoring weighted estimator, with treatment discontinuation defined as a ≥30-day gap.

Results: For all NCOs, women initiating weekly oral alendronate were more likely to discontinue treatment (range: 50-59%) compared to those initiating biannual denosumab injections (range: 20-28%). In ITT analyses, the risk among denosumab users was elevated for over half of the NCOs compared to alendronate users, suggesting residual bias. When censoring for treatment discontinuation, the risk of NCOs was similar between the denosumab and alendronate groups, consistent with the hypothesis that these groups are comparable dependent on measured covariates. For example, the RD (95% confidence interval) for having received an annual wellness exam was 3% (1%, 4%) in the ITT analysis versus 1% (-1%, 3%) in the analysis that censored for treatment discontinuation.

Conclusions: Our findings highlight that adherence to an OP biologic versus a small-molecule drug may be highly differential. Interestingly, our NCO analyses suggest that residual bias may nonetheless be negligible when censoring for treatment discontinuation, presumably because removing the high proportion of patients who discontinue oral alendronate renders that group more comparable to injectable denosumab users with respect to health-seeking behaviors and other unmeasured mechanisms of adherence.